Seite - 9 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 2226 (2) Select 0.5 popsize better chaotic disturbance fruit flies. Compute the fitness value of each Fruit fly from 2popsize chaotic disturbance fruit flies, and arrange these fruit flies to be a sequencebasedontheorderoffitnessvalues. Then, select the fruitflieswith0.5popsizeranking ahead in the fitness values; as a result, the 0.5 popsize better chaotic disturbance fruit flies areobtained. (3) Determine0.5popsizecurrent fruitflieswithbetterfitness. Compute thefitnessvalueofeach Fruit fly fromcurrentQFOA,andarrange these fruitflies tobeasequencebasedontheorderof fitnessvalues. Then, select the fruitflieswith0.5popsizerankingaheadin thefitnessvalues. (4) FormthenewCQFOApopulation.Mixthe0.5popsizebetterchaoticdisturbancefruitflieswith 0.5popsizecurrent fruitflieswithbetterfitness fromcurrentQFOA,andformanewpopulation thatcontainsnew1popsize fruitflies,andnameit thenewCQFOApopulation. (5) Completeglobalchaoticperturbation. AfterobtainingthenewpopulationofCQFOA, take it as thenewpopulationofQFOAandcontinuetoexecute theQFOAprocess. 2.2.4. ImplementationStepsofCQFOA Thestepsof theproposedCQFOAforparameteroptimizationofanLS-SVRmodelareas follows asshowninFigure1. Step1. Initialization. ThepopulationsizeofquantumDrosophila is1popsize; themaximumnumber of iterations isGen-max; therandomsearchradius isR; andthechaosdisturbancecontrol coefficient isNGCP. Step2. Randomsearching. For quantumrotationangle, θij, of a randomsearch, according to the quantumrotation angle, fruit fly locations on eachdimension are updated, and then, a quantum revolvingdoor isappliedtoupdate thequantumsequence,asshowninEquations(26)and(27) [34,35]: θij = θ(j)+R×rand(1) (26) qij= abs ([ cosθij −sinθij sinθij cosθij ] ×Q(j) ) , (27) where i is an individual of quantum fruit flies, i = 1,2,. . . ,1popsize; j is the position dimension of quantum fruit flies, j = 1,2,. . . , l. As mentioned above, the position of qij is non-negative constrained, thus, the absolute function, abs() is used to take the absolute value of each element in thecalculationresult. Step3.Calculatingfitness.MappingeachDrosophila location,qi, to thefeasibledomainofanLS-SVR model parameters to receive the parameters, (γi,σi). The training data are used to complete the trainingprocessesof theLS−SVRi modelandcalculate the forecastingvalue in the trainingstage correspondingtoeachsetofparameters. Then, the forecastingerror is calculatedas inEquation(12)of CQFOAbythemeanabsolutepercentageerror (MAPE),asshowninEquation(28): MAPE= 1 N N ∑ i=1 ∣∣∣∣∣ fi(x)− fˆi(x)fi(x) ∣∣∣∣∣×100%, (28) whereN is the totalnumberofdatapoints; fi(x) is theactual loadvalueatpoint i; and fˆi(x) is the forecasted loadvalueatpoint i. Step4.Choosingthecurrentoptimum. Calculate the tasteconcentrationof fruitfly,Smelli, byusing Equation(12),andfindthebestflavorconcentrationof individual,Best_Smelli,byEquation(13),as the optimalfitnessvalue. Step 5. Updating global optimization. Comparewhether the contemporary odor concentration, Best_Smelli=current, is better than the global optima, Best_Smelli. If so, update the global value by 9